Abstract:

Segmentation is a central task in Musical Analysis. We can say that it is the partition of a musical flow, usually following some set of criteria, into homogeneous segments. This paper introduces the Sonic Object Segmentator (SOS), a computer tool designed for this task. The theoretical model SOS is based on is Sonic Object Oriented Analysis. It considers the sonic object as the basic component of musical structure. It is defined as the result of the combination and interaction of some statistical components that describe distribution patterns of pitches in space and time. The model also asserts that a continuity break in at least one of these components strongly points out a new structurally significant unit. SOS' structure is based on Mathema, a multiagent system where artificial agents, each one with a specific knowledge about one of these components, and the human user cooperate in order to accomplish some task. SOS attempts to find, for a given musical work (MIDI formatted), the set of structurally most important components of the piece, and to give back the best segmentation, according to the observed continuity breaks in those components.